Type: Talk
Session: Machine Learning and Artificial Intelligence
Author: Shar Daniels

Abstract: With the novel application of deep learning models to continuously-exposed astronomical data, we are creating tools to discover and reveal the nature of rapidly-evolving optical astrophysical phenomena. Phenomena that vary on this timescale include cataclysmic variables, blazars, occultations by solar system objects, and potentially the optical counterpart of Fast Radio Bursts. Yet, the evolution of optical astronomical phenomena at sub-second timescales is under-explored due to technical limitations in traditional observing modes which require seconds-to-minutes exposure and readout cycles. However, two nontraditional observing modalities, trailing and continuous-readout, enable resolution at sub-second timescales by integrating the images of astrophysical objects along one spatial dimension. Analyses of these data require custom-made analysis pipelines. We are developing neural networks for the analysis of 450GB continuous-readout astronomical dataset sampled at 300 Hz from the Zwicky Transient Facility (ZTF). This poster will show the performance of CNN and transformer models under development on the ZTF data and outline the potential of our analysis tools for detecting and analyzing rapid transients at scale in this dataset and for LSST, for which trailing image modes have been suggested.